How the Banking Sector Leverages Predictive APIs

“The new trend is to describe a traditional business and then add the words ‘AI.’ A few years back, it was ‘social’.” ING Bank’s Natalino Busa isn’t wrong about the sudden fuss over artificial intelligence. But in certain sectors, adding AI into the mix indeed means revolutionizing it. Perhaps no more is this change happening than in the banking sector.

The recent PAPIs Connect brought together the academics and business people alike to talk about how predictive APIs are making artificial intelligence, machine learning and even deep learning more accessible. Yes, there’s the sci-fi end of the AI world happening, but there’s also a growing penetration of AI and machine learning for better customer service, particularly in the banking space, which was one of the focus tracks of the conference. Predictive APIs are allowing banks and alternative financial institutions to bridge the gap between the heavily scholastic side of the discipline and the customers and businesses that can benefit from it.

The Value of Big Data to Banks

“Data is the fabric of our lives. Data is at the relationship with your customer—it helps the customer with actionable data and an ethical approach,” Busa said. But how does it help the customer? He said data in banking can be used for:

  • proposing or suggesting
  • advising
  • offering choices
  • filtering
  • connecting
  • simplifying processes

At ING, an example of these initiatives has been rolled out to small business and personal banking customers in Poland and Spain, where a platform gives customers insights into personal finance in a very visual way. The Dutch bank is also researching using Spark Streaming and Strata Hadoop to create real-time fraud detection.

Busa added ING’s priority for data—and what he believes should be a professional priority for all banks—is to protect the customer by:

  • detecting
  • preventing
  • alerting
  • blocking
  • defending
  • identifying
  • authorizing

There’s no doubt that the massive amount of specific personally identifiable data out there growing rapidly in a scary way. Machine learning offers ways of anonymizing that data, while still constantly improving how it recognizes anomalies like potentially fraudulent behavior. The application programming interface or API allows different departments and banking branches to access and use this knowledge and to translate it in accessible, beneficial way for the clients. The predictive API and banking APIs in general allow banks and third-party developers to build consumer-friendly apps that leverage the power and knowledge derived from machine learning.

How the API is Bringing Machine Learning to the Banking Sector

The diagram above shows how ING has progressed with machine learning. Busa explained, “We are in a Classic Machine Learning approach where we are still hand-designing our features from input.”

But where does the API fit into this AI? Predictive APIs—the increasingly common term for the APIs that give broader access to machine learning—in the banking sector are also being used for:

  • feature extraction: extracting only the relevant information from a lot of data
  • anomaly detection: also called novelty detection or outlier detection, for data that doesn’t conform to expectations
  • identification based on behavior: finding patterns in how the data behaves
  • fraud detection: to find abnormal “red flag” behavior

APIs are currently the driving force behind the voice, speech and video recognition that will be used to better identify that you are really you, the owner of your bank account. Using the diagram above, APIs are common ways for data scientists to gain those Inputs from aggregate sources and for app developers to take the Outputs and turn them into useful tools.

But perhaps the most compelling use case for APIs across machine learning and artificial intelligence is that APIs are used to connect and translate among data scientists, engineers and domain experts. It’s as if data scientists that are hand-designing the learned features are speaking a different language—which often they are coding in the more complicated data-centric languages of R, Scala or Python—from the app designers, user experience designers, and other more customer-facing developers. APIs then allow this latter group access the benefits of machine learning without having to earn a PhD in it.

So if you’re looking to start leveraging predictive APIs to drive data science, Busa recommends you start lean, and focusing on the business proposition. Data science must deliver an outcome. It’s not just about delivering a model. It’s about asking: How can you quantify that model? This comes down to focusing on these questions:

  • What’s the improvement?
  • can you quantify your model quality?
  • can you quantify impact?
  • what is the impact for the customer journey?
  • can it be converted in financial results?

Busa analogues the API as the ladder that makes AI and machine learning easy to use, but that the data and science are the rails. “Provide the right data, provide the right science to cross the chasm,” to connect data, tech and business.

Be sure to read the next Banking article: Open Banking Goes Mainstream

 

Comments (0)